Understanding the social impacts of enforcement activities on illegal wildlife trade in China

Illegal wildlife trade enforcement is a cornerstone conservation strategy worldwide, yet we have a limited understanding on its social impacts. Using Chinese online wildlife seizure news (2003–2018), we evaluated the interactions among enforcement operations, news frequency, and social engagement (i.e., whistle-blowing) frequency. Our results showed that intensive enforcement operations, which commenced after 2012, have social impacts by increasing the frequency of all seizure news significantly by 28% [95% Confidence Interval (CI): 5%, 51%] and those via whistle-blowing by 24% [95% CI: 2%, 45%], when compared to counterfactual models where possible confounding factors were accounted for. Furthermore, we revealed the potential interaction between enforcement seizure news with and without social engagement, and the consequential social feedback process. Of the species identified from ‘whistle-blowing’ news, up to 28% are considered as high conservation priorities. Overall, we expanded our understanding of the enforcement impacts to social dimensions, which could contribute to improving the cost-effectiveness of such conservation efforts. Supplementary Information The online version contains supplementary material available at 10.1007/s13280-021-01686-9.

We repeatedly checked for duplication and news filtering protocol to remove unrelated news, including those without specific locality, topics about zoo and farms and those unofficial investigations by journalists, resulting in 2137 news left. We then further excluded 10 seizure news where Hong Kong, Macau or Taiwan was implicated either as an origin, a transit, or a destination. Based on another round of duplication check, we finalized a set of 2020 news for further analyses. For each unique and non-duplicated seizure news, the information extracted included the year of the seizure, the media source, the name of species with the quantity or weight and the location of the incidents (see the flow chart in Fig. S1).

Data analysis with equations
The Bayesian structural time series model (BSTS) The BSTS model allows us to evaluate the posterior inference of impacts and estimate the cumulative effect of the interventions on the number of seizure news reports over time. It could be defined as follows (Brodersen et al., 2015): The equation includes all the components that explain the number of the observed data . The first component, , is the value of the trend at time t. The component is the expected increase in , between t and t + 1, presents a random walk pattern. So, it can be thought of as the slope at time t. The vector is a set of potential control series candidate to be predictive of the response. All covariates are assumed to be contemporaneous. The error terms follow independent Gaussian random ).

Granger-causality Test
Given two sets of time series data, and , granger-causality is a method for determining whether one series is likely to provide more information about future values of the other than past values of itself alone (Granger, 1969;Granger, 1980). It is based on linear prediction theory for detecting causal relations among multiple linear time series, which is usually implemented via autoregressive (AR) modelling of multivariate time series. This is accomplished by taking different lags of one series and using that to model the change in the second series. It creates two models which predict y, one with only past values of and below where k is the number of lags in the time series:  To evaluate the sample completeness and coverage, sample-size, coverage-based rarefaction and extrapolation metrics were used to provide the estimates of total species (Chao and Jost, 2012;Chao et al., 2014). We performed the sample completeness curves (95% CI) in (Chao et al. 2014), which reveals sample completeness for a given sample size. This curve also provides an estimate of the sample size needed to achieve a certain degree of completeness.

Supplementary Results
The impact of enforcement on social engagement at the provincial level  Table S6).
In Yunnan province ( Interactive effects between different types of seizure cases at the provincial level At the provincial-level, granger causality tests for determining the interactive effects between the trends of normal news and whistle-blowing news could provide regional reference to decision-making of public outreach efforts in each province. Generally, stable effect that normal news could have influence on whistle-blowing news were detected for a certain duration after accumulated periods in Yunnan, Guangdong, Guangxi and Zhejiang, while this effect was seen earliest in Guangdong. In particular, the significant grangercausality was observed in Guangdong about one year after the intensive enforcement commenced in November 2013. Furthermore, the granger-causal relationship in Yunnan appeared slightly later than Guangdong in March 2015, followed by Zhejiang (observed in March 2016) and Guangxi (appeared in February 2017). In contrast, in Hunan and Anhui, the granger-causality of normal news on whistle-blowing news was strong at the beginning then decrease steadily (Fig. S3).
Specifically, in Guangdong province, we detected a stable effect where the normal seizure news significantly predicted whistle-blowing news after the intensive enforcement  Figure S1. The flow chart of the data collection and information extraction (To be continued in following page).

Key words searching on Baidu News
Time period = one month wildlife + seized / smuggling / poaching       3. The nationwide enforcement operation by Customs that non-specifically targeting wildlife are not in the list.
* presents the international joint operation for tackling wildlife smuggling.
2. Significance: * p < 0.05, ** p < 0.01, *** p < 0.001.   1. One seizure news could include more than one taxonomic group on class level. The overall number would be larger than the actual number of seizure news collected because of the overlapping.
2. There are 1476 seizure news that the information of taxonomic groups could be identified on species level.
3. Taxonomic groups on family level might cover more beyond the number of species, since some common name in news only can reach family level.